We introduce FormalMATH, a large-scale Lean4 benchmark for formal mathematical reasoning, comprising 5,560 semantically validated Lean4 statements spanning competition-level and undergraduate mathematics. To improve reliability, FormalMATH is constructed through a human-in-the-loop pipeline that combines type checking, LLM-based semantic verification against the informal–formal statements alignment, negation-based disproving, and expert review. We conduct a systematic evaluation of frontier theorem provers under test-time scaling on FormalMATH. Results show that current provers remain far from solving the benchmark: the strongest model, DeepSeek-Prover-V2 (671B), achieves only 28.31% Pass@32. We further observe substantial domain imbalance across provers, with stronger performance in applied mathematics and algebra but weaker results in discrete mathematics. We identify common failure modes in current provers, including incomplete proofs, weak handling of inequalities, misuse of automation tactics, and redundant hypothesis introduction. Extreme test-time scaling (e.g., Pass@3200) yields only modest gains on a subset of FormalMATH, suggesting that current training paradigms do not equip provers to effectively exploit additional sampling budget. We also observe a counterintuitive negative effect when provers are given human-written natural-language solution references: such references can measurably degrade formal proof generation rather than aid it. Overall, FormalMATH provides a challenging and reliable testbed for measuring progress in formal mathematical reasoning.

Reinforcement learning with verifiable rewards (RLVR) has become a central paradigm for improving reasoning and code generation in large language models, and GRPO-style training is widely adopted for its simplicity and effectiveness. However, an important design choice remains underexplored: how token-level policy gradient terms are aggregated within each sampled group. Standard GRPO uses sequence aggregation, while recent work has advocated token aggregation as a better alternative. We show that these two rules induce different optimization biases: token aggregation introduces sign-length coupling, while sequence aggregation implicitly downweights longer responses through sequence-level equal weighting. To address this tension, we propose Balanced Aggregation (BA), a simple drop-in replacement that computes token-level means separately within the positive and negative subsets and then combines them with sequence-count-based weights. Experiments with Qwen2.5-Math-7B and Qwen3-1.7B on DAPO-17k and Polaris, evaluated on six reasoning and coding benchmarks, show that BA consistently improves training stability and final performance over standard token and sequence aggregation. Our analysis further shows that the relative effectiveness of token and sequence aggregation is largely governed by response-length variation and the positive-negative length gap, highlighting aggregation as a critical design dimension in GRPO-style RLVR.

The rapid evolution of reasoning-intensive Large Language Models renders traditional metrics insufficient by masking fine-grained failures and implicit pathologies. Existing weakness discovery methods typically rely on rigid pipelines, yielding superficial insights that lack the diagnostic depth required for effective model improvement. To address this, we introduce Agent4Weakness, a multi-agent framework designed to replicate the rigorous workflow of human expert analysts. By integrating a Domain-Aware Memory for contextual reasoning with diagnostic evaluation skills and a Tool Abstraction mechanism for decouple data analysis, Agent4Weakness transforms raw evaluation traces into grounded, actionable reports. We validate our framework through an extensive study involving 104 models across 27 benchmarks. Experimental results demonstrate that Agent4Weakness produces diagnostic reports significantly superior to competitive baselines. Crucially, leveraging these insights for prompt guidance yields an average 3.7 point performance boost and establishes a closed-loop optimization framework.

Multilingual fluency often invites a stronger assumption: a model that can speak a user’s language must also understand the culture encoded by that language. We call this the Illusion of Cultural Alignment. To test this assumption directly, we introduce MMC-QA, a benchmark of 1,086 natively sourced questions across 11 language groups, five cultural dimensions, and three difficulty tiers. Unlike translated benchmarks, MMC-QA targets locally grounded knowledge and reduces shortcuts from English-centric cross-lingual transfer. Evaluating 15 LLMs, we find substantial cultural degradation and a pronounced Locality Effect: cultural competence tracks pre-training exposure more closely than general reasoning ability. We further show that common inference-time remedies do not dissolve the illusion. Models remain overconfident on unfamiliar cultural questions, repeated sampling yields unstable rather than reliable correctness, and retrieval augmentation helps unevenly on long-tail facts. These findings indicate that cultural alignment cannot be inferred from multilingual ability alone and requires deeper intervention than calibration, sampling, or retrieval at inference time.

Existing Chinese preference datasets suffer from limited scale, restricted domain coverage, and insufficiently rigorous data validation. Human annotation significantly limits the scalability of human preference datasets. As a result, Chinese Alignment and Chinese Reward Models (CRM) have not yet been thoroughly explored. To address these challenges, we design an LLM-based data annotation pipeline with no human intervention. Based on this pipeline, we curate COIG-P (Chinese Open Instruction Generalist - Preference), a high-quality, large-scale Chinese preference dataset consisting of 1M Chinese preference pairs and 92k carefully curated Chinese queries across diverse domains, including Chat, Coding, Maths, and others. We conduct experiments to verify the quality of COIG-P from two perspectives. (1) COIG-P brings significant performance improvements for the Qwen2/2.5 and Infinity-Instruct model series on AlignBench through DPO, with gains ranging from 2% to 12%. Furthermore, it significantly outperforms other existing Chinese preference datasets. (2) We train an 8B-sized CRM and manually annotate a Chinese Reward Benchmark (CRBench). Our CRM demonstrates robust scoring ability on CRBench. In addition, in practical data construction experiments, the quality of the data constructed by our CRM is comparable to that produced by GPT-4o.

Current multi-modal benchmarks primarily focus on facts within individual images. However, they overlook the associative relations among multiple images, which necessitate conducting commonsense reasoning grounded in associated knowledge at different granularities (i.e., image-level and entity-level) as well as the ability to perceive the order of images. Therefore, we propose a multi-image relational association task and a meticulously curated Multi-granularity Multi-image Relational Association (MMRA) benchmark, comprising 1,024 samples. To systematically evaluate current LVLMs, we establish a system of associative relations among images that contains 11 subtasks (e.g., UsageSimilarity, SubEvent, etc.) at two granularity levels (i.e., image-level and entity-level), based on relations in ConceptNet. Our experiments reveal that entity-level multi-image perception tasks pose greater challenges for LVLMs than image-level tasks. Moreover, LVLMs perform poorly on spatial-related tasks, indicating limited spatial awareness. Furthermore, we find that LVLMs exhibit weak image order perception capabilities, and we design a method to significantly improve this ability, demonstrating that most current LVLMs do not adequately consider image order perception during pre-training.

Large Reasoning Models (LRMs) have demonstrated strong performance by producing extended Chain-of-Thought (CoT) traces before answering. However, this paradigm often induces over-reasoning: redundant calculations and circular self-verification that increase computational cost without improving outcomes. Existing evaluations largely emphasize final accuracy or coarse token counts, and lack automated tools to separate essential logic from structural redundancy. We introduce CoTJudger, a graph-driven framework that quantifies reasoning efficiency by converting free-form CoTs into directed dependency graphs and extracting the Shortest Effective Path (SEP) needed to reach a correct solution. This yields an interpretable efficiency signal – how much of a CoT is necessary versus structurally redundant – that is comparable across models and tasks. Evaluating 21 LRMs, CoTJudger reveals pervasive redundancy and surfaces recurring failure modes, including verification obsession and compensatory redundancy. These results provide a practical metric for disentangling reasoning ability from computational waste, enabling more targeted evaluation and diagnosis of LRM efficiency.

Large Reasoning Models (LRMs) benefit substantially from training on challenging competition-level questions. However, existing automated question synthesis methods lack precise difficulty control, incur high computational costs, and struggle to generate competition-level questions at scale. In this paper, we propose CoDiQ (Controllable Difficult Question Generation), a novel framework enabling fine-grained difficulty control via test-time scaling while ensuring question solvability. Specifically, first, we identify a test-time scaling tendency (extended reasoning token budget boosts difficulty but reduces solvability) and the intrinsic properties defining the upper bound of a model’s ability to generate valid, high-difficulty questions. Then, we develop CoDiQ-Generator from Qwen3-8B, which improves the upper bound of difficult question generation, making it particularly well-suited for challenging question construction. Building on the CoDiQ framework, we build CoDiQ-Corpus (44K competition-grade question sequences). Human evaluations show these questions are significantly more challenging than LiveCodeBench/AIME with over 82% solvability. Training LRMs on CoDiQ-Corpus substantially improves reasoning performance, verifying that scaling controlled-difficulty training questions enhances reasoning capabilities. We open-source CoDiQ-Corpus, CoDiQ-Generator, and implementations to support related research.

Large Language Models (LLMs) perform well on standard reasoning and question-answering benchmarks, yet such evaluations often fail to capture their ability to handle long-tail, expertise-intensive knowledge in real-world professional scenarios. We introduce LPFQA, a long-tail knowledge benchmark derived from authentic professional forum discussions, covering 7 academic and industrial domains with 430 curated tasks grounded in practical expertise. LPFQA evaluates specialized reasoning, domain-specific terminology understanding, and contextual interpretation, and adopts a hierarchical difficulty structure to ensure semantic clarity and uniquely identifiable answers. Experiments on over multiple mainstream LLMs reveal substantial performance gaps, particularly on tasks requiring deep domain reasoning, exposing limitations overlooked by existing benchmarks. Overall, LPFQA provides an authentic and discriminative evaluation framework that complements prior benchmarks and informs future LLM development. Our data is available at https://huggingface.co/datasets/m-a-p/LPFQA.

AI agents with advanced reasoning and tool-use capabilities have demonstrated impressive performance in web browsing for deep search. However, existing benchmarks such as BrowseComp primarily focus on textual content, overlooking the prevalence of multimodal content. To bridge this gap, we introduce MM-BrowseComp, a novel benchmark comprising 400 challenging, hand-crafted questions designed to evaluate multimodal retrieval and reasoning capabilities. Unlike prior work, MM-BrowseComp incorporates visual prompts and necessitates the extraction of key evidence from web images and videos to complete questions, rendering text-only approaches insufficient. Additionally, we provide a verified checklist for each question, enabling fine-grained analysis of multimodal dependencies and reasoning paths. Our comprehensive evaluation of 27 state-of-the-art models reveals that even leading models like GPT-5-High with tools achieve only 24.25% accuracy, highlighting the suboptimal multimodal browsing capabilities, establishing MM-BrowseComp as a rigorous new standard for the field.

Translating natural language mathematical statements into formal, executable code is a fundamental challenge in automated theorem proving. While prior work has focused on generation and compilation success, little attention has been paid to the critic phase—the evaluation of whether generated formalizations truly capture the semantic intent of the original problem. In this paper, we introduce CriticLean, a novel critic-guided reinforcement learning framework that elevates the role of the critic from a passive validator to an active learning component. Specifically, first, we propose the CriticLeanGPT, trained via supervised fine-tuning and reinforcement learning, to rigorously assess the semantic fidelity of Lean 4 formalizations. Then, we introduce CriticLeanBench, a benchmark designed to measure models’ ability to distinguish semantically correct from incorrect formalizations, and demonstrate that our trained CriticLeanGPT models can significantly outperform strong open- and closed-source baselines. Building on the CriticLean framework, we construct FineLeanCorpus, a dataset comprising over 509K problems that exhibits rich domain diversity, broad difficulty coverage, and high correctness based on human evaluation.Overall, our findings highlight that optimizing the critic phase is essential for producing reliable formalizations and we hope our CriticLean will provide valuable insights for future advances in formal mathematical reasoning.

Remarkable progress on large language models (LLMs), particularly in English, has facilitated impressive capabilities in following human instructions. However, there remains a noticeable gap in instruction fine-tuning for Chinese, where the complex linguistic features pose significant challenges. Existing datasets, generally distilled from English-centric LLMs, are not well-aligned with Chinese users’ interaction patterns. To bridge this gap, we introduce COIG-CQIA, a new Chinese instruction tuning dataset derived from various real-world data resources and undergoing comprehensive human verification. We conduct extensive experiments on COIG-CQIA, and compare them with strong baseline models and datasets. The experimental results show that models trained on COIG-CQIA achieve highly competitive performance in diverse benchmarks. Additionally, our findings offer several insights for designing effective Chinese instruction-tuning datasets and data mixing strategies. Our dataset are available at https://huggingface.co/datasets/m-a-p/COIG-CQIA.

Large Language Models (LLMs), e.g. ChatGPT, have been widely adopted in real-world dialogue applications. However, LLMs’ robustness, especially in handling long complex dialogue sessions, including frequent motivation transfer, sophisticated cross-turn dependency, is criticized all along. Nevertheless, no existing benchmarks can fully reflect these weaknesses. We present MARS-Bench, a Multi-turn Athletic Real-world Scenario Dialogue Benchmark, designed to remedy the gap. MARS-Bench is constructed from play-by-play text commentary so to feature realistic dialogues specifically designed to evaluate three critical aspects of multi-turn conversations: ultra multi-turn, interactive multi-turn, and cross-turn tasks. Extensive experiments on MARS-Bench also reveal that closed-source LLMs significantly outperform open-source alternatives, explicit reasoning significantly boosts LLMs’ robustness on handling long complex dialogue sessions, and LLMs indeed face significant challenge when handling motivation transfer and sophisticated cross-turn dependency. Moreover, we provide mechanistic interpretability on how attention sinks due to special tokens lead to LLMs’ performance degradation when handling long complex dialogue sessions based on attention visualization experiment in Qwen2.5-7B-Instruction.

Multimodal Large Language Models (MLLMs) are measured on numerous benchmarks like image captioning, visual question answer, and reasoning. However, these benchmarks often include overly simple or uninformative samples, making it difficult to effectively distinguish the performance of different MLLMs. Additionally, evaluating models across many benchmarks creates a significant computational burden. To address these issues, we propose LIME (Less Is More for MLLM Evaluation), a refined and efficient benchmark curated using a semi-automated pipeline. This pipeline filters out uninformative samples and eliminates answer leakage by focusing on tasks that require image-based understanding. Our experiments show that LIME reduces the number of samples by 76% and evaluation time by 77%, while it can more effectively distinguish different models’ abilities. Notably, we find that traditional automatic metrics like CIDEr are insufficient for evaluating MLLMs’ captioning performance, and excluding the caption task score yields a more accurate reflection of overall model performance. All code and data are available at https://anonymous.4open.science/r/LIME-49CD

In this paper, we introduce MIO, a novel foundation model built on multimodal tokens, capable of understanding and generating speech, text, images, and videos in an end-to-end, autoregressive manner. While the emergence of large language models (LLMs) and multimodal large language models (MM-LLMs) propels advancements in artificial general intelligence through their versatile capabilities, they still lack true any-to-any understanding and generation. Recently, the release of GPT-4o has showcased the remarkable potential of any-to-any LLMs for complex real-world tasks, enabling omnidirectional input and output across images, speech, and text. However, it is closed-source and does not support the generation of multimodal interleaved sequences. To address this gap, we present MIO, which is trained on a mixture of discrete tokens across four modalities using causal multimodal modeling. MIO undergoes a four-stage training process: (1) alignment pre-training, (2) interleaved pre-training, (3) speech-enhanced pre-training, and (4) comprehensive supervised fine-tuning on diverse textual, visual, and speech tasks. Our experimental results indicate that MIO exhibits competitive, and in some cases superior, performance compared to previous dual-modal baselines, any-to-any model baselines, and even modality-specific baselines. Moreover, MIO demonstrates advanced capabilities inherent to its any-to-any feature, such as interleaved video-text generation, chain-of-visual-thought reasoning, visual guideline generation, instructional image editing, etc.

As the capabilities of Multimodal Large Language Models (MLLMs) improve, the need for higher-order evaluation of them is increasing. However, there is a lack of work evaluating MLLM for higher-order perception and understanding of Chinese visual content. To address this, we introduce the CII-Bench, which aims to assess MLLMs’ such capabilities for Chinese images. To ensure the authenticity of the Chinese context, images in CII-Bench are sourced from the Chinese Internet and manually reviewed, with corresponding answers also manually crafted. Additionally, CII-Bench incorporates images that represent Chinese traditional culture, such as famous Chinese traditional paintings, which can deeply reflect the model’s understanding of Chinese traditional culture. Through experiments on multiple MLLMs using CII-Bench, significant findings emerged. There is a large gap between MLLMs and humans in performance. The highest MLLM accuracy is 64.4%, while the human average is 78.2% and the peak is 81.0%. MLLMs perform poorly on traditional culture images, indicating limitations in understanding high-level semantics and lacking a deep knowledge base of Chinese traditional culture. Moreover, most models have higher accuracy when image emotion hints are added to the prompts. We believe CII-Bench will help MLLMs better understand Chinese semantics and specific images, and move forward the development of expert artificial general intelligence (AGI). Our project is publicly available at https://cii-bench.github.io.

Alignment of large language models (LLMs) involves training models on preference-contrastive output pairs to adjust their responses according to human preferences. To obtain such contrastive pairs, traditional methods like RLHF and RLAIF rely on limited contrasting patterns, such as varying model variants or decoding temperatures. This singularity leads to two issues: (1) alignment is not comprehensive; and thereby (2) models are susceptible to harmful response tendencies. To address these issues, we investigate how to construct more comprehensive and diversified contrasting patterns to enhance preference data (RQ1) and verify the impact of the diversification of contrasting patterns on model alignment (RQ2). For RQ1, we propose PopAlign, a framework that integrates diversified contrasting patterns across the prompt, model, and pipeline levels, introducing six contrasting strategies that do not require additional feedback labeling procedures. Regarding RQ2, we conduct thorough experiments demonstrating that PopAlign significantly outperforms existing methods, leading to more comprehensive alignment.

Large Language Models (LLMs) demonstrate impressive capabilities across various domains, including role-playing, creative writing, mathematical reasoning, and coding. Despite these advancements, LLMs still encounter challenges with length control, frequently failing to adhere to specific length constraints due to their token-level operations and insufficient training on data with strict length limitations. We identify this issue as stemming from a lack of positional awareness and propose novel approaches—PositionID Prompting and PositionID Fine-Tuning—to address it. These methods enhance the model’s ability to continuously monitor and manage text length during generation. Additionally, we introduce PositionID CP Prompting to enable LLMs to perform copy and paste operations accurately. Furthermore, we develop two benchmarks for evaluating length control and copy-paste abilities. Our experiments demonstrate that our methods significantly improve the model’s adherence to length constraints and copy-paste accuracy without compromising response quality.

The advent of Large Language Models (LLMs) has paved the way for complex tasks such as role-playing, which enhances user interactions by enabling models to imitate various characters. However, the closed-source nature of state-of-the-art LLMs and their general-purpose training limit role-playing optimization. In this paper, we introduce RoleLLM, a framework to benchmark, elicit, and enhance role-playing abilities in LLMs. RoleLLM comprises four stages: (1) Role Profile Construction for 100 roles; (2) Context-Based Instruction Generation (Context-Instruct) for role-specific knowledge extraction; (3) Role Prompting using GPT (RoleGPT) for speaking style imitation; and (4) Role-Conditioned Instruction Tuning (RoCIT) for fine-tuning open-source models along with role customization. By Context-Instruct and RoleGPT, we create RoleBench, the first systematic and fine-grained character-level benchmark dataset for role-playing with 168,093 samples. Moreover, RoCIT on RoleBench yields RoleLLaMA (English) and RoleGLM (Chinese), significantly enhancing role-playing abilities and even achieving comparable results with RoleGPT (using GPT-4).

Multi-modal information retrieval (MMIR) is a rapidly evolving field where significant progress has been made through advanced representation learning and cross-modality alignment research, particularly in image-text pairing.However, current benchmarks for evaluating MMIR performance on image-text pairings overlook the scientific domain, which has a notable gap with the generic data since the caption of scientific charts and tables usually describes the analysis of experimental results or scientific principles in contrast to human activity or scenery depicted in generic images.To bridge this gap, we develop a scientific domain-specific MMIR benchmark (SciMMIR) by leveraging open-access research paper corpora to extract data relevant to the scientific domain. This benchmark comprises 530K meticulously curated image-text pairs, extracted from figures and tables with detailed captions from scientific documents.We further annotate the image-text pairs with a two-level subset-subcategory hierarchy to facilitate a more comprehensive evaluation of the baselines. We conduct zero-shot and fine-tuned evaluations on prominent multi-modal image-captioning and visual language models, such as CLIP, BLIP, and BLIP-2.Our findings offer critical insights for MMIR in the scientific domain, including the impact of pre-training and fine-tuning settings and the effects of different visual and textual encoders.

The advancement of large language models (LLMs) has enhanced the ability to generalize across a wide range of unseen natural language processing (NLP) tasks through instruction-following.Yet, their effectiveness often diminishes in low-resource languages like Chinese, exacerbated by biased evaluations from data leakage, casting doubt on their true generalizability to new linguistic territories. In response, we introduce the Chinese Instruction-Following Benchmark (CIF-Bench), designed to evaluate the zero-shot generalizability of LLMs to the Chinese language. CIF-Bench comprises 150 tasks and 15,000 input-output pairs, developed by native speakers to test complex reasoning and Chinese cultural nuances across 20 categories. To mitigate data contamination, we release only half of the dataset publicly, with the remainder kept private, and introduce diversified instructions to minimize score variance, totaling 45,000 data instances.Our evaluation of 28 selected LLMs reveals a noticeable performance gap, with the best model scoring only 52.9%, highlighting the limitations of LLMs in less familiar language and task contexts.This work not only uncovers the current limitations of LLMs in handling Chinese language tasks but also sets a new standard for future LLM generalizability research, pushing towards the development of more adaptable, culturally informed, and linguistically diverse models.

While LLMs demonstrate impressive capabilities in musical knowledge, we find that music reasoning is still an unsolved task.We introduce ChatMusician, an open-source large language model (LLM) that integrates intrinsic musical abilities. It is based on continual pre-training and finetuning LLaMA2 on a text-compatible music representation, ABC notation, and the music is treated as a second language.ChatMusician can understand and generate music with a pure text tokenizer without external multi-modal neural structures or tokenizers. Interestingly, endowing musical abilities does not harm language abilities, even achieving a slightly higher MMLU score.ChatMusician is capable of composing well-structured, full-length music, condition on texts, chords, melodies, motifs, musical forms, etc.On our meticulously curated college-level music understanding benchmark, MusicTheoryBench, ChatMusician surpasses LLaMA2 and GPT-3.5 by a noticeable margin. We show that ChatMusician preserves or even surpasses the original LLaMA2 7B’s language abilities by evaluating on MMLU benchmark.Our work reveals that LLMs can be an excellent compressor for music, which can be seen as humanity’s creative language, but there remains significant territory to be conquered.We release our 5B token music-language corpora MusicPiles, the collected MusicTheoryBench, code, model and demo.

Machine Translation (MT) has developed rapidly since the release of Large Language Models and current MT evaluation is performed through comparison with reference human translations or by predicting quality scores from human-labeled data. However, these mainstream evaluation methods mainly focus on fluency and factual reliability, whilst paying little attention to figurative quality. In this paper, we investigate the figurative quality of MT and propose a set of human evaluation metrics focused on the translation of figurative language. We additionally present a multilingual parallel metaphor corpus generated by post-editing. Our evaluation protocol is designed to estimate four aspects of MT: Metaphorical Equivalence, Emotion, Authenticity, and Quality. In doing so, we observe that translations of figurative expressions display different traits from literal ones.